A Neural Network-Based Modeling Method for Terminal Charges in GaN HEMTs

Abstract

In this paper, the detailed description of neural network (NN)-based modeling method for terminal charges in gallium nitride high electron mobility transistor (GaN HEMT) is presented. This method helps to overcome originally high modeling difficulty of terminal charges while their modeling accuracy can be guaranteed. A two-hidden layer ANN is constructed to model the source-end terminal charge (qis) and drain-end terminal charge ($q$id) in the MIT Virtual Source (MVS) theory, respectively. The MVS theory is used to connect the ANN-based terminal charges and drain current. This method is benchmarked with accurate drain current performance.